Specify hyperparameters manually

You usually have a few configurations worth trying first — sensible defaults, or settings that worked on a similar problem. OptunaSharp gives you two ways to feed them in:

Study.EnqueueTrial — hand OptunaSharp the parameter sets and let it evaluate them (before it starts sampling on its own).
Study.AddTrial — register sets you have already evaluated, so the sampler takes their results into account without re-running them.

Both use the parameter names below with a placeholder objective; substitute your real training code.

First scenario: have OptunaSharp evaluate your hyperparameters

EnqueueTrial queues a configuration to be evaluated next. Queued trials run before sampled ones, so your defaults are tried first:

using OptunaSharp;
using OptunaSharp.Collections;
using OptunaSharp.Pruners;

double Objective(Trial trial)
{
    double baggingFraction  = trial.SuggestFloat("bagging_fraction", 0.4, 1.0 + 1e-12);
    long   baggingFreq      = trial.SuggestInt("bagging_freq", 0, 7);
    long   minChildSamples  = trial.SuggestInt("min_child_samples", 5, 100);
    return TrainAndScore(baggingFraction, baggingFreq, minChildSamples);
}

var study = Optuna.CreateStudy(direction: StudyDirection.Maximize, pruner: new MedianPruner());

// Try the defaults, then a variant with a larger bagging frequency, before sampling.
study.EnqueueTrial(new OrderedDict<string, object?>
{
    ["bagging_fraction"] = 1.0, ["bagging_freq"] = 0, ["min_child_samples"] = 20,
});
study.EnqueueTrial(new OrderedDict<string, object?>
{
    ["bagging_fraction"] = 0.75, ["bagging_freq"] = 5, ["min_child_samples"] = 20,
});

study.Optimize(Objective, nTrials: 100, timeout: 600);

Second scenario: reuse already-evaluated hyperparameters

If you have results for those sets, register them as completed trials with AddTrial. The sampler then models them as history — no re-evaluation. Build each completed trial with FrozenTrial.CreateTrial, supplying the @params, their distributions, and the observed value:

using OptunaSharp;
using OptunaSharp.Collections;
using OptunaSharp.Distributions;
using OptunaSharp.Pruners;

var study = Optuna.CreateStudy(direction: StudyDirection.Maximize, pruner: new MedianPruner());

var distributions = new OrderedDict<string, BaseDistribution>
{
    ["bagging_fraction"]  = new FloatDistribution(0.4, 1.0 + 1e-12),
    ["bagging_freq"]      = new IntDistribution(0, 7),
    ["min_child_samples"] = new IntDistribution(5, 100),
};

study.AddTrial(FrozenTrial.CreateTrial(
    value: 0.94,
    @params: new OrderedDict<string, object?>
    {
        ["bagging_fraction"] = 1.0, ["bagging_freq"] = 0L, ["min_child_samples"] = 20L,
    },
    distributions: distributions));

study.AddTrial(FrozenTrial.CreateTrial(
    value: 0.95,
    @params: new OrderedDict<string, object?>
    {
        ["bagging_fraction"] = 0.75, ["bagging_freq"] = 5L, ["min_child_samples"] = 20L,
    },
    distributions: distributions));

study.Optimize(Objective, nTrials: 100, timeout: 600);

CreateTrial validates that every parameter lies inside its declared distribution, so a typo in a bound fails fast rather than corrupting the study.

Relationship to Optuna

Optuna (Python)	OptunaSharp
`study.enqueue_trial({"bagging_freq": 5, …})`	`study.EnqueueTrial(new OrderedDict<string, object?> { ["bagging_freq"] = 5, … })`
`optuna.trial.create_trial(params=…, distributions=…, value=0.94)`	`FrozenTrial.CreateTrial(@params: …, distributions: …, value: 0.94)`
`optuna.distributions.FloatDistribution(0.4, 1.0 + 1e-12)`	`new FloatDistribution(0.4, 1.0 + 1e-12)`
`optuna.distributions.IntDistribution(0, 7)`	`new IntDistribution(0, 7)`
`study.add_trial(trial)`	`study.AddTrial(trial)`

Table of Contents

Specify hyperparameters manually

First scenario: have OptunaSharp evaluate your hyperparameters

Second scenario: reuse already-evaluated hyperparameters

Relationship to Optuna

See also